How Anthropic Built Claude Code
AI, tech, startup and product podcast recommendations for today
The AI, tech, startup and product podcast recommendations for today are:
Boris Cherny, creator of Claude Code at Anthropic, shares how building for models six months ahead boosted engineer productivity 70% and now powers 4% of public commits.
Anto pitches Cosmic Brain AI on The Pitch, unveiling VR glasses that let anyone train robots from video demos, slashing engineering costs with a skills marketplace.
Google principal engineer Taylor Mullen explains shipping 150 Gemini CLI features weekly by granting AI agents full OS access while enforcing strict human oversight.
Kavak CEO Carlos Garcia Otati discusses rebuilding his company around AI agents that now handle 90% of customer interactions after enduring initial performance dips.
Limitless Podcast rigorously tests OpenClaw across user types, spotlighting FelixBot’s $41,000 weekly earnings alongside critical security risks and safer alternatives.
Pick an episode below, press play, lace up and get moving.
Podcast Recommendations
1. How Anthropic Built Claude Code (w/ Boris Cherny)
Y Combinator
Listen on: Spotify | Apple Podcasts | Podlink (50 m)
Topics: AI | Product | Productivity | Programming | Technology
Boris Cherny, creator of Claude Code at Anthropic, reveals how building for future models—not today’s—transformed coding productivity by 150% and may eliminate the software engineer job title entirely.
● Build for the model six months ahead, not today’s capabilities, to avoid being leapfrogged by competitors.
● Claude Code’s terminal design emerged accidentally from latent user demand, not a master plan.
● Anthropic’s productivity per engineer grew 70% despite doubling team size; 4% of all public commits now use Claude Code.
“Don’t build for the model of today, build for the model six months from now.” - Boris Cherny
The Pitch
Listen on: Spotify | Apple Podcasts | Podlink (41 m)
Topics: AI | Product | Robotics | Startups | Venture Capital
Anto pitches Cosmic Brain AI, software that democratises robot training via VR glasses and video, enabling anyone to teach robots new skills without millions in engineering costs.
● Vision Language Action models translate human video into robot motor movements, enabling one-shot learning from single demonstrations.
● Marketplace model lets developers sell trained robot skills as APIs, generating revenue per robot usage hour across any hardware brand.
● B2B focus on airlines, automotive, and warehouses proves faster traction than consumer robotics; six-month follow-up shows pre-seed closed at $1M.
“We are building the cursor for robotics. This is the future of robot training.” - Anto
3. How Google’s Gemini CLI Creator Ships 150 Features a Week
The Neuron: AI Explained
Listen on: Spotify | Apple Podcasts | Podlink (56 m)
Topics: AI | Product | Productivity | Programming
Taylor Mullen, Google principal engineer and Gemini CLI creator, reveals how his team ships 100–150 features weekly by giving AI agents full OS access instead of confining them to text editors.
● Gemini CLI enables agentic search through codebases using grep and file reading rather than embeddings, mimicking how human developers solve problems.
● Policy files and hard-coded permission checks prevent agents from executing unauthorised actions, maintaining human oversight at every critical step.
● The Ralph Wiggum technique—refeeding outputs into fresh contexts iteratively—refines AI solutions and enables longer autonomous workflows without derailing.
“Right now, out of the default on our team, everyone’s 10X. So we’re not thinking about better anymore, because the 10X is almost the new normal.” - Taylor Mullen
4. From Copilots to Agents: Rebuilding the Company Around AI
a16z Podcast
Listen on: Spotify | Apple Podcasts | Podlink (1 h)
Topics: AI | Automation | Leadership | Product
Kavak CEO Carlos Garcia Otati reveals replacing copilots with AI agents handling 90% of customer interactions after a painful year of flat growth.
● Build ontology first for agent communication.
● Deploy agents on hardest funnels initially.
● Endure performance dips for AI superiority.
“When you’re building AI, the first thing that you need to build is the brakes of the system.” - Carlos Garcia Otati
5. Inside OpenClaw: Should You Actually Use It? Probably Not.
Limitless Podcast
Listen on: Spotify | Apple Podcasts | Podlink (23 m)
Topics: AI | Automation | Cybersecurity | Product | Technology
Hosts test OpenClaw’s real-world capabilities across three user archetypes, revealing which use cases deliver value and which require patience.
● FelixBot generated $41,000 weekly through autonomous book creation, token trading, and skill marketplace transactions.
● Security risks include brute-force attacks, credential exposure, and prompt injection vectors requiring sandboxed deployment.
● Claude Cowork offers safer alternative; wait for OpenAI and Anthropic’s curated versions launching within three to six months.
“Imagine giving ChatGPT access to your wallet, your medical records, and allowing it to run loose in the world.”